When it comes to security, LLMs are like Swiss cheese — and that’s going to cause huge problems
We recorded 7 claims from this post. 3 were checked against evidence. 4 can’t be checked. 4 more are held. Read the post at its source.
Claims from this post (7)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: UnresolvedSaid
“The more people use LLMs, the more trouble we are going to be in.”
What we tested: The more people use LLMs, the more trouble we are going to be in.
Assessment: SupportedSaid
“Jailbreaks aren’t new, but even after years of them, the tech industry has nothing like a robust response.”
What we tested: Despite years of jailbreaks, the tech industry has no robust defense against them.
What happened: The study evaluates four defenses and reports substantial vulnerability to human multi-turn attacks despite strong results against automated single-turn attacks. This supports the ordinary reading that established defenses lacked broad robustness. (arXiv; Scale AI and UC Berkeley researchers, )
Assessment: Can’t be checkedSaid
“In hindsight, ChatGPT may come to be seen as the greatest publicity stunt in AI history, an intoxicating glimpse at a future that may actually take years to realize—kind of like a 2012-vintage driverless car demo, but this time with a foretaste of an ethical guardrail that will take years to perfect.”
What we tested: ChatGPT may come to be seen as AI's greatest publicity stunt, and its ethical guardrails will take years to perfect.
Assessment: Can’t be checkedSaid
“Nothing that has happened since changes that view.”
What we tested: Marcus says nothing since his February 2023 essay has changed his view that ChatGPT may come to be seen as the greatest publicity stunt in AI history, a glimpse of a future that may take years to realize, with ethical guardrails that will take years to perfect.
Assessment: Can’t be checkedSaid
“The possibilities are now endless for propaganda, troll farms, and rings of fake websites that degrade trust across the internet.”
What we tested: LLM chatbots open wide possibilities for propaganda, troll farms and fake websites that erode trust online.
Assessment: UnresolvedSaid
“And, bonus, there is little if any government regulation in place to do much about this.”
What we tested: There is little if any government regulation in place to address chatbot guardrail failures.
Assessment: Can’t be checkedSaid
“It’s a disaster in the making.”
What we tested: The state of LLM chatbot safety is a disaster in the making.
Held from this post (4)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews reached different assessments. Our reading, not assessed: Marcus reaffirms as still true his February 2023 summary that ChatGPT, the world's most used chatbot, is governed by undisclosed training data and a barely described algorithm, is glorified by the media, and has ethical guardrails that only sorta kinda work and are driven more by text similarity than by any true moral calculus.
- The two reviews read the claim differently
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: Companies such as Google, Tesla and Figure.AI are integrating LLMs that are vulnerable to jailbreaks into robots.